Re: Deep Learning is the problem
"or you just program them to avoid all objects in the road that are bipedal."
That's a good start. Now what about women pushing bikes across the road? Or pushchairs/prams with babies in them? Animals? How do we teach it to avoid a bottle dropped or blown into the road? Most humans would likely spot if it was plastic and not worry too much, or glass and try to avoid it. When will it be good enough to distinguish between potential hazards and large numbers of autumn leaves either blowing in the wind or getting caught in a small (or large) moving pile in a wind vortex? As I posted yesterday, this is really really hard stuff to program for. I'd be very interested to see each of the leading AV makers test their cars on a windy Autumn day down a tree lined avenue. Not to mention in snow where all the lane markings are covered over.
Biting the hand that feeds IT
